Is it safe to put customer details into ChatGPT?

What happens to a customer's name and order when you paste them into ChatGPT or Claude, the settings to check, and how to get the same help without them.

17 October 2026 · The Workden team

It's ten at night and a customer has sent you a long, cross email about a parcel that arrived crushed. You're tired, you want to reply well, so you copy the whole thread into ChatGPT and type "help me answer this kindly". Her name goes in with it. So does her address, her order number and the bit where she mentions she's buying it for her mum's birthday.

The reply it writes is lovely. Then, halfway through sending it, you wonder where all that went.

The short answer: it depends less on the AI than on which account you're signed in to, and two settings you can check in a minute. On a personal plan, assume what you paste could be kept and possibly used to train future models unless you've switched that off. On a business plan, the default is that it isn't used for training. And for most shop jobs, you don't need to paste the customer's details at all. The AI writes just as good an apology for "the customer" as it does for Mrs Patel.

Two kinds of account, two different deals

The same ChatGPT app can sit on very different terms depending on the plan.

On the business side, OpenAI's enterprise privacy page covers ChatGPT Business, Enterprise and its API, and says plainly: "We do not train our models on your data by default." Anthropic says the same of its business products, Claude for Work and its API, on its commercial training page.

Personal plans work differently. In ChatGPT, OpenAI's data controls page describes a setting called Improve the model for everyone. With it off, your new conversations won't be used to train their models. For Claude Free, Pro and Max, Anthropic's consumer page says chats are used to improve its models if you choose to allow it, which is a setting in your privacy settings.

If you signed up for ChatGPT or Claude at home, you're on a personal plan. That's fine for writing product descriptions. It's the one to be careful with when a customer's details are involved.

The two settings worth checking tonight

In ChatGPT: Settings, then Data controls, then turn off Improve the model for everyone. OpenAI's page says this applies across your devices when you're signed in. If you want a chat to vanish from your history, use Temporary chat, which OpenAI says isn't used to improve its models but may be kept for up to 30 days for safety.

In Claude: open your privacy settings and look at Model Improvement. Anthropic also offers Incognito chats, which it says are not used to improve Claude even if that setting is on.

The button nobody thinks about

This is the part that catches people out. When you click thumbs up or thumbs down on an answer, you're sending feedback, and feedback carries the whole conversation with it.

OpenAI's data controls page says that if you give feedback after opting out of training, "the entire conversation associated with that feedback may be used to train OpenAI models." Anthropic says it will store the entire related conversation from feedback for up to five years, on both its consumer and business plans.

So if you pasted a customer's email and then gave the reply a thumbs up out of habit, that customer's email went with it. If there's anything personal in a chat, leave the buttons alone.

The better habit: leave the customer out

One idea makes most of this go away. The AI doesn't need to know who the customer is to help you. It needs to know what happened and how you want to sound.

Before you paste, swap out anything that points to a real person:

Name becomes "the customer"
Order number becomes "their order"
Address, phone and email come out entirely
Anything personal they mentioned (a birthday, an illness, a house move) goes in only if it changes the reply, and then in general words

Then ask in your own words. Something like this works well:

A customer's parcel arrived crushed. It was a gift and she's upset because it was meant for a birthday this weekend. We can send a replacement by next-day post at our cost. Write a short, warm reply that apologises without grovelling and tells her what happens next. Sound like a small independent shop, not a big company.

You'll get a reply as good as the one with her name in it. Add the name and the order number yourself when you paste it into your email.

This takes about thirty seconds, and it's the habit that keeps you on the right side of your customers and your own privacy notice.

What your privacy notice says

If you're in the UK, your shop's privacy notice is supposed to tell customers who you share their details with. The ICO's guidance on what privacy information to provide says to "Say who you share people's personal data with", and that this includes anyone who processes it on your behalf.

That doesn't mean you can never use an AI tool. It means that if you plan to put customer details into one regularly, it should be a business account, and your privacy notice should mention it. If you're unsure whether your notice covers it, that's a question for whoever wrote it. If you're in the US, the rules depend on where you and your customers are, so check what applies to you.

If you only ever paste anonymised text, as above, there's no customer's personal data going in, and the question mostly answers itself.

What this doesn't cover

This post is about the everyday chat apps. It doesn't cover AI features built into your shop platform, your email tool or your accounts software, which each have their own terms. Look for the provider's own privacy or data page, not a blog summary of it, and check whether they train on what you give them.

Where Workden fits

Sometimes you do want the AI to know the customer, because the useful answer depends on their history. "Has this person ordered before? What did we promise them last time?" Copying all that into a chat by hand is exactly the pasting we've been warning about.

In Workden, you connect your own ChatGPT or Claude and ask in plain words. It answers from your customer records and drafts the reply for you, and nothing is sent or changed until a person approves it. The connection is made from your AI's own settings, never with its password, and you can switch it off in one click. The settings on your AI account still apply, so the checks above are worth doing either way.

Workden is £10.99 a month for your whole team, with a 14 day free trial. If you'd like more on writing replies to upset customers, how to get fewer "where is my order" emails is a good place to start.

Questions shop owners ask

Is ChatGPT Plus private?

Plus is a personal plan. Whether your chats help train OpenAI's models depends on the Improve the model for everyone setting, so check it. The business plans are the ones where training is off by default.

You don't need to. Write the newsletter with the AI, then send it from your email tool to your list. The AI never needs to see who's on it.

Is Temporary chat the same as private?

Not quite. OpenAI says temporary chats aren't used to improve its models and don't appear in your history, but may be kept for up to 30 days for safety.

Sources

Checked 17 October 2026.